To Save the Economy... We’re About to Break It
The Fed is raising rates, but the economy's biggest growth engine—AI—is built on debt. Can it survive?
The Federal Reserve's recent decision to raise interest rates for the first time in three years marked a significant shift in monetary policy. This unanimous vote by the Federal Open Market Committee (FOMC) followed a period of considerable division, with a 9-to-3 vote against rate hikes just seven weeks prior. The change in sentiment, as explained by Fed Chair Kevin Worsh, was driven by inflation data indicating that price increases were not abating independently and that existing monetary policy was insufficient to curb excess demand.
This policy adjustment has coincided with a surge in the 10-year Treasury yield, a benchmark for numerous other interest rates, which has surpassed 5%, reaching its highest point since 2007. Historically, Federal Reserve tightening cycles, of which there have been 11 since 1965, typically involve multiple rate increases over an extended period. The average cycle has lasted approximately 18 months and added close to 4% to rates. Only one prior cycle, in March 1997 under Alan Greenspan, consisted of a single rate hike before a subsequent reduction in rates 18 months later. Projections from 16 of 18 Fed officials suggest at least one more rate hike is anticipated before the end of the current year. The Conference Board forecasts three consecutive hikes in September, October, and December.
Research analyzing 14 tightening cycles between 1955 and 2009 by the New York Fed revealed that 10 were followed by a recession within 18 months of the final hike. Of the remaining four, one led to a significant increase in unemployment and a credit crunch, which some economists consider a recession. This leaves only three instances in 50 years where the U.S. economy achieved a "soft landing"—raising interest rates without triggering a recession.
Traditionally, Federal Reserve rate hikes have impacted the economy through the housing market. Increased mortgage rates made home purchases and refinancing less attractive, leading to reduced construction and consumer spending, thereby cooling the economy. However, this mechanism appears to be weakening. Approximately 40% of U.S. households hold a mortgage, and 78% of these are at rates below 6%, with two-thirds below 5% and half below 4%. This "rate lock" phenomenon, where homeowners are disinclined to sell due to significantly lower existing mortgage rates compared to current market rates, has led to a frozen housing market. Existing home sales are currently around 4 million annually, below the 15-year average of approximately 5 million. Last year marked the slowest sales pace since 1995.
With the housing market's traditional role as a monetary policy lever diminishing, the impact of rate hikes is shifting. Fed Chair Kevin Worsh highlighted the substantial capital flowing into AI-related infrastructure as a key area. JP Morgan estimates that the global data center buildout could cost approximately $5 trillion by 2030. A finance professor at Columbia University calculated that the U.S. AI buildout, if realized as planned, could account for roughly 2.8% of the nation's annual economic output through 2032. This scale surpasses historical infrastructure projects like the transcontinental railroad (2.5% of the economy in the 1800s), nationwide electrification (1%), and the interstate highway system (1.5%).
While tech companies are central to this buildout, their internal cash flows, stock sales, and generated revenue only cover about a quarter of the projected 4 trillion will be financed through borrowing. This has led to a significant increase in bond issuance by major tech companies. From 2020 to 2024, the five leading tech companies involved in the AI boom issued an average of 93 billion in 2025. By July of the current year, issuance had already reached 830 billion, representing half the volume of the entire U.S. investment-grade bond market, two-thirds of all leveraged loans, and more than the entire high-yield bond market combined. Vanguard projects capital spending by these companies to exceed $1 trillion annually over the next three years.
The cost of borrowing has become a critical factor. When companies issue bonds, they pay the government borrowing rate plus a "spread," which reflects market nervousness about repayment. This year, the spread on the top five tech companies has widened by approximately 30 basis points, 15 times the increase seen in the broader investment-grade market. Debt financing data center projects is priced about 100 basis points wider, and junk-rated debt is around 200 basis points wider.
The situation is particularly acute for companies at the lower end of the AI market. A 267 million to $640 million. In the second quarter, interest payments consumed 42 cents of every dollar of adjusted earnings for CoreWeave, indicating a real-time repricing at the bottom of the AI market.
Mercatus, a compute pricing firm, modeled the profitability of a single GPU cluster at various costs of capital. At 6%, it is profitable; at 10%, it barely breaks even; and at 12%, it becomes unprofitable, with a break-even point around 11%. The firm categorizes buyers by their cost of capital: established companies pay 6-8%, late-stage AI companies 10-14%, and earlier-stage venture-backed companies 15-20%. This means the top of the market clears the break-even, the middle is at the line, and the bottom is already underwater. Borrowing costs are not static; floating-rate debt, like the CoreWeave loan priced off an overnight rate plus 5.5%, can increase rapidly with Fed rate hikes. Fixed-rate debt also resets upon maturity, with companies rolling over debt at prevailing market rates.
The average tightening cycle since 1983 has added over 3 percentage points to rates. With 16 of 18 Fed officials anticipating at least one more hike this year and The Conference Board forecasting three, the cost of debt for the AI buildout is expected to rise. This contrasts with previous tightening cycles where the impact was primarily on consumer demand. This cycle's significant impact is on the supply side, affecting companies with substantial debt loads undertaking massive construction projects. The risk is that when these companies need to refinance or seek additional capital, the increased cost of borrowing could destabilize the buildout.
The AI buildout is currently growing at approximately 0.85 percentage points of GDP annually, a rate faster than the housing boom (0.5% annually) and the telecom buildout of the 1990s (0.15% annually). This rapid expansion, which is projected to contribute nearly 1.5 percentage points to U.S. economic growth this year, carries echoes of the housing bubble. Private data center construction has surged by 57% in one year to $75 billion annually, more than doubling in two years. Data from the Census Bureau indicates that U.S. spending on data center construction now exceeds spending on conventional office buildings.
While no immediate signs of spending pullback or slowdown are evident, JP Morgan's analysis of the top five tech companies reveals a dramatic shift. Capital spending, which historically never exceeded 13% of revenue, is projected to reach 41% this year, nearly consuming all incoming revenue. This results in negative free cash flow, necessitating increased debt financing at rising interest rates. This dynamic connects directly to the Fed's recent rate hike decision, as the largest capital investment in American history will increasingly face higher borrowing costs. This situation, often overlooked in traditional media, highlights the critical challenge of financing a massive, ongoing buildout in a rising interest rate environment.
The Fed's Unanimous Decision and Inflation Concerns
The Federal Reserve's shift from a divided stance in July to a unanimous decision to raise interest rates in September highlights a change in their assessment of inflation. Fed Chair Kevin Worsh cited persistent inflation and insufficient monetary policy to curb excess demand as key drivers for the rate hike.
- Federal Reserve raised interest rates for the first time in three years.
- The decision was unanimous (12-0), contrasting with a 9-3 vote in July.
- In July, 9 officials voted to leave rates unchanged, while 3 voted to raise them.
- Fed Chair Kevin Worsh stated inflation was 'not abating on its own' and policy was insufficient.
- Inflation was described as 'not going away' and rates were not high enough to slow spending.
Historical Context of Rate Hikes and Market Reaction
The bond market reacted to the Fed's decision, with the 10-year Treasury yield surpassing 5%, a level not seen since 2007. Historical analysis of Fed tightening cycles shows that most involve multiple hikes and significant rate increases, with few stopping after a single hike. The Fed's own projections indicate further hikes are likely.
- The 10-year Treasury yield closed above 5%, the highest since 2007.
- Historically, Fed tightening cycles (since 1965) typically involve multiple hikes over about 1.5 years, adding close to 4%.
- The smallest historical rate increase was 1.75 percentage points; the largest was 13%.
- Only one past tightening cycle (March 1997 under Alan Greenspan) stopped after a single hike.
- Fed projections show 16 of 18 officials expect at least one more hike before year-end.
- The Conference Board forecasts three rate hikes in September, October, and December.
The High Risk of Recession Following Rate Hikes
Historical data reveals a high probability of recession or economic downturn following Fed tightening cycles. Out of 14 cycles studied between 1955-2009, 10 were followed by a recession within 18 months of the final hike. Of the remaining four, one led to a significant jump in unemployment and a credit crunch, leaving only three instances of a 'soft landing' in 50 years.
- 10 out of 14 tightening cycles (1955-2009) were followed by a recession within 18 months of the final hike.
- One of the remaining four cycles resulted in a jump in unemployment and a credit crunch.
- Only three 'soft landings' (raising rates without crashing the economy) occurred in 50 years.
- The impact of rate hikes depends on how the cycle progresses, not just the number of hikes.
The Weakening Impact of Rate Hikes on the Housing Market
The traditional impact of rate hikes on the housing market, a key economic lever, is diminishing. With a majority of households locked into low mortgage rates, fewer people are selling homes, leading to a frozen housing market. This means the Fed's usual method of slowing the economy is becoming less effective.
- Approximately 40% of American households have a mortgage.
- 78% of those with mortgages are locked in below 6%.
- Two-thirds are below 5%, and half are under 4%.
- High current rates (near 7%) discourage homeowners with low rates from selling.
- This phenomenon is termed 'rate lock', leading to reduced housing market activity.
- Existing home sales are at 4 million/year, below the 15-year average of 5 million.
- Last year was the slowest for housing sales since 1995.
- The housing market is described as 'already frozen'.
The AI Buildout: A New Driver of Debt and Economic Activity
The massive AI infrastructure buildout, estimated to cost trillions, is becoming the new engine of borrowing and economic activity. This unprecedented scale of investment, driven by tech companies, is heavily reliant on the bond market, creating significant demand for debt and potentially exposing the economy to new risks as borrowing costs rise.
- Fed Chair Kevin Walsh previously highlighted capital flowing into AI infrastructure.
- Global data center buildout is estimated to cost roughly $5 trillion through 2030 (JP Morgan).
- US AI buildout could represent 2.8% of annual US GDP through 2032.
- This scale rivals or exceeds historical infrastructure projects like railroads or the interstate highway system.
- Tech companies' operating cash flow covers only about a quarter of the $5.5 trillion cost.
- An estimated $4 trillion needs to be borrowed.
- Bond issuance by five major tech companies increased from 93 billion in 2025.
- AI infrastructure spend is projected to exceed $830 billion this year (Chadam Financial).
- This spend represents half the volume of the US investment-grade bond market.
- Vanguard projects capital spending by these companies above $1 trillion annually for the next 3 years.
Rising Borrowing Costs and Their Impact on AI Infrastructure
The cost of borrowing for AI infrastructure is rising significantly, with spreads widening considerably for tech companies compared to the broader market. Debt for data center projects is priced much higher, and even profitable companies are facing increased interest costs and stricter loan conditions, pushing some projects towards unprofitability.
- The price of money is central to the AI buildout story.
- Spreads on bonds for five major tech companies widened by 30 basis points this year, 15 times the move in the broader investment-grade market.
- Debt for data center projects prices about 100 basis points wider than corporate bonds.
- Junk territory debt prices about 200 basis points wider.
- Coreweave faced a $2.6 billion loan syndication that was returned; interest rates increased by a full point.
- New loan conditions for Coreweave included full repayment over life, coverage tests, and minimum cash balance.
- Coreweave's interest bill increased from 640 million.
- Interest costs consumed 42 cents of every dollar of adjusted earnings for Coreweave in Q2.
The Break-Even Point for AI Projects and Rising Costs
The profitability of AI compute projects is highly sensitive to the cost of capital. Modeling shows that at higher borrowing costs (around 11-12%), these projects become unprofitable. While established companies may still clear break-even, later-stage and early-stage AI companies face significant challenges as borrowing costs continue to rise, resetting higher with each loan maturity.
- A single GPU cluster becomes unprofitable at a cost of capital around 11-12% (Mercatis model).
- Established companies pay 6-8% on new deals.
- Late-stage AI companies pay 10-14%.
- Early-stage venture-backed AI companies pay 15-20%.
- Borrowing costs reset periodically, with floating rates adjusting quickly to Fed hikes.
- Fixed-rate debt will reset higher upon maturity and rollover.
- The average tightening cycle since 1983 added over 3 percentage points to borrowing costs.
Supply-Side Impact: How This Cycle Differs
Unlike previous tightening cycles that primarily impacted consumer demand, the current cycle's main effect is on the supply side, specifically on companies undertaking massive debt-financed projects like the AI buildout. This shift means that the consequences of rate hikes may not be immediate but will manifest later when companies need to refinance or seek new funding at higher costs.
- The current tightening cycle impacts the supply side, unlike past cycles that hit demand (mortgages, credit cards, car loans).
- The primary impact is on companies with large debt loads building future infrastructure.
- Rate hikes don't break things immediately; they cause problems later when refinancing or seeking new capital at higher costs.
- Massive demand for debt is colliding with a tightening cycle.
- AI buildout is growing roughly twice as fast as the housing boom (2002-2005).
- AI capex will provide a 1.5 percentage point boost to US economic growth this year.
- Private data center construction is up 57% year-over-year, more than doubled in two years.
The Unprecedented Scale and Financial Strain of the AI Buildout
The AI buildout's rapid growth and immense capital requirements are unprecedented, surpassing historical booms like housing or telecom. Tech companies are dedicating an increasingly large portion of their revenue to capital spending, leading to negative free cash flow and a greater reliance on debt financing at rising costs. This situation connects directly to the Fed's recent rate hike decision, as the economy's growth is now heavily dependent on this debt-fueled expansion.
- AI buildout growth rate is 0.85% of GDP annually, faster than the housing boom.
- AI buildout is the largest capital investment in American history.
- AI capex will boost US economic growth by nearly 1.5 percentage points this year.
- Private data center construction spending is $75 billion/year, up 57% in one year.
- US spends more on data centers than conventional office buildings.
- For the top five tech companies, capital spending rose from under 13% of revenue to 41% this year.
- Free cash flow for these companies has turned negative for the first time.
- Future financing will increasingly rely on debt at rising rates.
- The Fed's rate hikes are directly linked to the need for continued debt financing for the AI buildout.